It's currently the amba/platform driver's responsibility to initialize the
pointer, dma_parms, for its corresponding struct device. The benefit with this
approach allows us to avoid the initialization and to not waste memory for the
struct device_dma_parameters, as this can be decided on a case by case basis.
However, it has turned out that this approach is not very practical. Not only
does it lead to open coding, but also to real errors. In principle callers of
dma_set_max_seg_size() doesn't check the error code, but just assumes it
succeeds.
For these reasons, this series initializes the dma_parms from the amba/platform
bus at the device registration point. This also follows the way the PCI devices
are being managed, see pci_device_add().
If it turns out that this is an acceptable solution, we probably also want the
changes for stable, but I am not sure if it applies without conflicts.
The series is based on v5.6.
Kind regards
Ulf Hansson
Ulf Hansson (2):
driver core: platform: Initialize dma_parms for platform devices
amba: Initialize dma_parms for amba devices
drivers/amba/bus.c | 1 +
drivers/base/platform.c | 2 ++
include/linux/amba/bus.h | 1 +
include/linux/platform_device.h | 1 +
4 files changed, 5 insertions(+)
--
2.20.1
It's currently the amba driver's responsibility to initialize the pointer,
dma_parms, for its corresponding struct device. The benefit with this
approach allows us to avoid the initialization and to not waste memory for
the struct device_dma_parameters, as this can be decided on a case by case
basis.
However, it has turned out that this approach is not very practical. Not
only does it lead to open coding, but also to real errors. In principle
callers of dma_set_max_seg_size() doesn't check the error code, but just
assumes it succeeds.
For these reasons, let's do the initialization from the common amba bus at
the device registration point. This also follows the way the PCI devices
are being managed, see pci_device_add().
Suggested-by: Christoph Hellwig <[email protected]>
Cc: Russell King <[email protected]>
Cc: <[email protected]>
Signed-off-by: Ulf Hansson <[email protected]>
---
Changes in v2:
- Move initialization to amba_device_initialize() to be more consistent
with how we manage platform devices.
---
drivers/amba/bus.c | 1 +
include/linux/amba/bus.h | 1 +
2 files changed, 2 insertions(+)
diff --git a/drivers/amba/bus.c b/drivers/amba/bus.c
index fe1523664816..8558b629880b 100644
--- a/drivers/amba/bus.c
+++ b/drivers/amba/bus.c
@@ -645,6 +645,7 @@ static void amba_device_initialize(struct amba_device *dev, const char *name)
dev->dev.release = amba_device_release;
dev->dev.bus = &amba_bustype;
dev->dev.dma_mask = &dev->dev.coherent_dma_mask;
+ dev->dev.dma_parms = &dev->dma_parms;
dev->res.name = dev_name(&dev->dev);
}
diff --git a/include/linux/amba/bus.h b/include/linux/amba/bus.h
index 26f0ecf401ea..0bbfd647f5c6 100644
--- a/include/linux/amba/bus.h
+++ b/include/linux/amba/bus.h
@@ -65,6 +65,7 @@ struct amba_device {
struct device dev;
struct resource res;
struct clk *pclk;
+ struct device_dma_parameters dma_parms;
unsigned int periphid;
unsigned int cid;
struct amba_cs_uci_id uci;
--
2.20.1
It's currently the platform driver's responsibility to initialize the
pointer, dma_parms, for its corresponding struct device. The benefit with
this approach allows us to avoid the initialization and to not waste memory
for the struct device_dma_parameters, as this can be decided on a case by
case basis.
However, it has turned out that this approach is not very practical. Not
only does it lead to open coding, but also to real errors. In principle
callers of dma_set_max_seg_size() doesn't check the error code, but just
assumes it succeeds.
For these reasons, let's do the initialization from the common platform bus
at the device registration point. This also follows the way the PCI devices
are being managed, see pci_device_add().
Suggested-by: Christoph Hellwig <[email protected]>
Cc: <[email protected]>
Signed-off-by: Ulf Hansson <[email protected]>
---
Changes in v2:
- Move initialization to setup_pdev_dma_masks(). This means the
initialization is done also in the OF path.
---
drivers/base/platform.c | 2 ++
include/linux/platform_device.h | 1 +
2 files changed, 3 insertions(+)
diff --git a/drivers/base/platform.c b/drivers/base/platform.c
index b5ce7b085795..c81b68d5d66d 100644
--- a/drivers/base/platform.c
+++ b/drivers/base/platform.c
@@ -361,6 +361,8 @@ struct platform_object {
*/
static void setup_pdev_dma_masks(struct platform_device *pdev)
{
+ pdev->dev.dma_parms = &pdev->dma_parms;
+
if (!pdev->dev.coherent_dma_mask)
pdev->dev.coherent_dma_mask = DMA_BIT_MASK(32);
if (!pdev->dev.dma_mask) {
diff --git a/include/linux/platform_device.h b/include/linux/platform_device.h
index 041bfa412aa0..81900b3cbe37 100644
--- a/include/linux/platform_device.h
+++ b/include/linux/platform_device.h
@@ -25,6 +25,7 @@ struct platform_device {
bool id_auto;
struct device dev;
u64 platform_dma_mask;
+ struct device_dma_parameters dma_parms;
u32 num_resources;
struct resource *resource;
--
2.20.1
> -----Original Message-----
> From: Ulf Hansson <[email protected]>
> Sent: 2020??4??1?? 2:39
> To: Greg Kroah-Hartman <[email protected]>; Rafael J . Wysocki
> <[email protected]>; [email protected]
> Cc: Arnd Bergmann <[email protected]>; Christoph Hellwig <[email protected]>;
> Russell King <[email protected]>; Linus Walleij
> <[email protected]>; Robin Murphy <[email protected]>; Vinod
> Koul <[email protected]>; BOUGH CHEN <[email protected]>; Ludovic
> Barre <[email protected]>; [email protected];
> [email protected]; Ulf Hansson <[email protected]>;
> [email protected]
> Subject: [PATCH v2 1/2] driver core: platform: Initialize dma_parms for
> platform devices
>
> It's currently the platform driver's responsibility to initialize the pointer,
> dma_parms, for its corresponding struct device. The benefit with this
> approach allows us to avoid the initialization and to not waste memory for
> the struct device_dma_parameters, as this can be decided on a case by case
> basis.
>
> However, it has turned out that this approach is not very practical. Not
> only does it lead to open coding, but also to real errors. In principle callers of
> dma_set_max_seg_size() doesn't check the error code, but just assumes it
> succeeds.
>
> For these reasons, let's do the initialization from the common platform bus
> at the device registration point. This also follows the way the PCI devices are
> being managed, see pci_device_add().
>
> Suggested-by: Christoph Hellwig <[email protected]>
> Cc: <[email protected]>
> Signed-off-by: Ulf Hansson <[email protected]>
>
Hi Ulf,
I test both on our local 5.4.24 branch and the latest Linux-next, both works fine.
Tested-by: Haibo Chen <[email protected]>
Best Regards
Haibo Chen
> ---
>
> Changes in v2:
> - Move initialization to setup_pdev_dma_masks(). This means the
> initialization is done also in the OF path.
>
> ---
> drivers/base/platform.c | 2 ++
> include/linux/platform_device.h | 1 +
> 2 files changed, 3 insertions(+)
>
> diff --git a/drivers/base/platform.c b/drivers/base/platform.c index
> b5ce7b085795..c81b68d5d66d 100644
> --- a/drivers/base/platform.c
> +++ b/drivers/base/platform.c
> @@ -361,6 +361,8 @@ struct platform_object {
> */
> static void setup_pdev_dma_masks(struct platform_device *pdev) {
> + pdev->dev.dma_parms = &pdev->dma_parms;
> +
> if (!pdev->dev.coherent_dma_mask)
> pdev->dev.coherent_dma_mask = DMA_BIT_MASK(32);
> if (!pdev->dev.dma_mask) {
> diff --git a/include/linux/platform_device.h
> b/include/linux/platform_device.h index 041bfa412aa0..81900b3cbe37
> 100644
> --- a/include/linux/platform_device.h
> +++ b/include/linux/platform_device.h
> @@ -25,6 +25,7 @@ struct platform_device {
> bool id_auto;
> struct device dev;
> u64 platform_dma_mask;
> + struct device_dma_parameters dma_parms;
> u32 num_resources;
> struct resource *resource;
>
> --
> 2.20.1
- Greg, Arnd, Linus, etc,
On Tue, 31 Mar 2020 at 20:38, Ulf Hansson <[email protected]> wrote:
>
> It's currently the amba/platform driver's responsibility to initialize the
> pointer, dma_parms, for its corresponding struct device. The benefit with this
> approach allows us to avoid the initialization and to not waste memory for the
> struct device_dma_parameters, as this can be decided on a case by case basis.
>
> However, it has turned out that this approach is not very practical. Not only
> does it lead to open coding, but also to real errors. In principle callers of
> dma_set_max_seg_size() doesn't check the error code, but just assumes it
> succeeds.
>
> For these reasons, this series initializes the dma_parms from the amba/platform
> bus at the device registration point. This also follows the way the PCI devices
> are being managed, see pci_device_add().
>
> If it turns out that this is an acceptable solution, we probably also want the
> changes for stable, but I am not sure if it applies without conflicts.
>
> The series is based on v5.6.
>
> Kind regards
> Ulf Hansson
>
>
> Ulf Hansson (2):
> driver core: platform: Initialize dma_parms for platform devices
> amba: Initialize dma_parms for amba devices
>
> drivers/amba/bus.c | 1 +
> drivers/base/platform.c | 2 ++
> include/linux/amba/bus.h | 1 +
> include/linux/platform_device.h | 1 +
> 4 files changed, 5 insertions(+)
>
> --
> 2.20.1
>
Does this look okay or is there anything you would like me to change?
Kind regards
Uffe
On Tue, Mar 31, 2020 at 8:38 PM Ulf Hansson <[email protected]> wrote:
>
> It's currently the platform driver's responsibility to initialize the
> pointer, dma_parms, for its corresponding struct device. The benefit with
> this approach allows us to avoid the initialization and to not waste memory
> for the struct device_dma_parameters, as this can be decided on a case by
> case basis.
>
> However, it has turned out that this approach is not very practical. Not
> only does it lead to open coding, but also to real errors. In principle
> callers of dma_set_max_seg_size() doesn't check the error code, but just
> assumes it succeeds.
>
> For these reasons, let's do the initialization from the common platform bus
> at the device registration point. This also follows the way the PCI devices
> are being managed, see pci_device_add().
>
> Suggested-by: Christoph Hellwig <[email protected]>
> Cc: <[email protected]>
> Signed-off-by: Ulf Hansson <[email protected]>
Reviewed-by: Arnd Bergmann <[email protected]>
On Tue, Mar 31, 2020 at 8:38 PM Ulf Hansson <[email protected]> wrote:
>
> It's currently the amba driver's responsibility to initialize the pointer,
> dma_parms, for its corresponding struct device. The benefit with this
> approach allows us to avoid the initialization and to not waste memory for
> the struct device_dma_parameters, as this can be decided on a case by case
> basis.
>
> However, it has turned out that this approach is not very practical. Not
> only does it lead to open coding, but also to real errors. In principle
> callers of dma_set_max_seg_size() doesn't check the error code, but just
> assumes it succeeds.
>
> For these reasons, let's do the initialization from the common amba bus at
> the device registration point. This also follows the way the PCI devices
> are being managed, see pci_device_add().
>
> Suggested-by: Christoph Hellwig <[email protected]>
> Cc: Russell King <[email protected]>
> Cc: <[email protected]>
> Signed-off-by: Ulf Hansson <[email protected]>
> ---
Reviewed-by: Arnd Bergmann <[email protected]>